dictionary["<PAD/>"] = default_vector
# models.append(dictionary)
return norm_model, raw_model
with Timer("lex"):
norm_model, raw_model = load_lexicon_unigram()
with Timer("w2v"):
w2vmodel = load_w2v()
unigram_lexicon_model = raw_model
# unigram_lexicon_model = raw_model
x_train, y_train, x_lex_train = cnn_data_helpers.load_data(
'trn', w2vmodel, unigram_lexicon_model, max_len)
x_dev, y_dev, x_lex_dev = cnn_data_helpers.load_data('dev', w2vmodel,
unigram_lexicon_model,
max_len)
x_test, y_test, x_lex_test = cnn_data_helpers.load_data(
'tst', w2vmodel, unigram_lexicon_model, max_len)
# x_train, y_train = cnn_data_helpers.load_data('trn',w2vmodel , max_len)
# x_dev, y_dev = cnn_data_helpers.load_data('dev', w2vmodel, max_len)
# x_test, y_test = cnn_data_helpers.load_data('tst', w2vmodel, max_len)
del (w2vmodel)
del (norm_model)
del (raw_model)
gc.collect()
print("Train/Dev split: {:d}/{:d}".format(len(y_train), len(y_dev)))
if self.name:
print '[%s]' % self.name,
print 'Elapsed: %s' % (time.time() - self.tstart)
def load_w2v():
model_path = '/Users/bong/works/data/word2vec_twitter_model/word2vec_twitter_model.bin'
with Timer("load w2v"):
model = Word2Vec.load_word2vec_format(model_path, binary=True)
print("The vocabulary size is: " + str(len(model.vocab)))
return model
w2vmodel = load_w2v()
x_train, y_train = cnn_data_helpers.load_data('trn',w2vmodel , max_len)
x_dev, y_dev = cnn_data_helpers.load_data('dev', w2vmodel, max_len)
# x_test, y_test = cnn_data_helpers.load_data('tst', w2vmodel, max_len)
del(w2vmodel)
gc.collect()
print("Train/Dev split: {:d}/{:d}".format(len(y_train), len(y_dev)))
# Training
# ==================================================
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
"Save model after this many steps (default: 100)")
# Misc Parameters
tf.flags.DEFINE_boolean("allow_soft_placement", True,
"Allow device soft device placement")
tf.flags.DEFINE_boolean("log_device_placement", False,
"Log placement of ops on devices")
FLAGS = tf.flags.FLAGS
FLAGS._parse_flags()
print("\nParameters:")
for attr, value in sorted(FLAGS.__flags.items()):
print("{}={}".format(attr.upper(), value))
print("")
max_len = 60
x_train, y_train, vocabulary, vocabulary_inv = cnn_data_helpers.load_data(
'trn', max_len)
x_test, y_test, _, _ = cnn_data_helpers.load_data('tst', max_len, vocabulary,
vocabulary_inv)
# savepath = './models/model-8200'
# savepath = './models/model-13800'
# savepath = './models/model-3900'
# savepath = './models/model-5050'
savepath = './models/model-6750'
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default():
cnn = TextCNN(sequence_length=x_train.shape[1],
tf.flags.DEFINE_integer("test_every", 614, "Evaluate model on test set after this many steps (default: 100)")
tf.flags.DEFINE_integer("checkpoint_every", 200, "Save model after this many steps (default: 100)")
# Misc Parameters
tf.flags.DEFINE_boolean("allow_soft_placement", True, "Allow device soft device placement")
tf.flags.DEFINE_boolean("log_device_placement", False, "Log placement of ops on devices")
FLAGS = tf.flags.FLAGS
FLAGS._parse_flags()
print("\nParameters:")
for attr, value in sorted(FLAGS.__flags.items()):
print("{}={}".format(attr.upper(), value))
print("")
max_len = 60
x_train, y_train, vocabulary, vocabulary_inv = cnn_data_helpers.load_data('trn', max_len)
x_test, y_test, _, _ = cnn_data_helpers.load_data('tst', max_len, vocabulary, vocabulary_inv)
# savepath = './models/model-8200'
# savepath = './models/model-13800'
# savepath = './models/model-3900'
# savepath = './models/model-5050'
savepath = './models/model-6750'
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default():
cnn = TextCNN(
sequence_length=x_train.shape[1],
def run_train(w2vdim, lexdim, lexnumfilters, sample_test = True):
if sample_test == True:
print '======================================[sample test]======================================'
max_len = 60
with Timer("lex"):
norm_model, raw_model = load_lexicon_unigram(lexdim)
with Timer("w2v"):
w2vmodel = load_w2v(w2vdim, sample_test=sample_test)
unigram_lexicon_model = norm_model
# unigram_lexicon_model = raw_model
if sample_test:
x_train, y_train, x_lex_train = cnn_data_helpers.load_data('trn_sample',w2vmodel, unigram_lexicon_model, max_len)
x_dev, y_dev, x_lex_dev = cnn_data_helpers.load_data('dev_sample', w2vmodel, unigram_lexicon_model, max_len)
x_test, y_test, x_lex_test = cnn_data_helpers.load_data('tst_sample', w2vmodel, unigram_lexicon_model, max_len)
else:
x_train, y_train, x_lex_train = cnn_data_helpers.load_data('trn', w2vmodel, unigram_lexicon_model, max_len)
x_dev, y_dev, x_lex_dev = cnn_data_helpers.load_data('dev', w2vmodel, unigram_lexicon_model, max_len)
x_test, y_test, x_lex_test = cnn_data_helpers.load_data('tst', w2vmodel, unigram_lexicon_model, max_len)
# x_train, y_train = cnn_data_helpers.load_data('trn',w2vmodel , max_len)
# x_dev, y_dev = cnn_data_helpers.load_data('dev', w2vmodel, max_len)
# x_test, y_test = cnn_data_helpers.load_data('tst', w2vmodel, max_len)
del(w2vmodel)
del(norm_model)
del(raw_model)
gc.collect()
print("Train/Dev split: {:d}/{:d}".format(len(y_train), len(y_dev)))
# Training
# ==================================================
if not FLAGS.random_seed:
tf.set_random_seed(FLAGS.seed_number)
with tf.Graph().as_default():
max_af1_dev = 0
index_at_max_af1_dev = 0
af1_tst_at_max_af1_dev = 0
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default():
if not FLAGS.random_seed:
tf.set_random_seed(FLAGS.seed_number)
cnn = TextCNNAttentionSimpleUT(
sequence_length=x_train.shape[1],
num_classes=3,
embedding_size=w2vdim,
embedding_size_lex=lexdim,
num_filters_lex = lexnumfilters,
filter_sizes=list(map(int, FLAGS.filter_sizes.split(","))),
num_filters=FLAGS.num_filters,
l2_reg_lambda=FLAGS.l2_reg_lambda)
# Define Training procedure
global_step = tf.Variable(0, name="global_step", trainable=False)
optimizer = tf.train.AdamOptimizer(1e-3)
grads_and_vars = optimizer.compute_gradients(cnn.loss)
train_op = optimizer.apply_gradients(grads_and_vars, global_step=global_step)
# Keep track of gradient values and sparsity (optional)
grad_summaries = []
for g, v in grads_and_vars:
if g is not None:
grad_hist_summary = tf.histogram_summary("{}/grad/hist".format(v.name), g)
sparsity_summary = tf.scalar_summary("{}/grad/sparsity".format(v.name), tf.nn.zero_fraction(g))
grad_summaries.append(grad_hist_summary)
grad_summaries.append(sparsity_summary)
grad_summaries_merged = tf.merge_summary(grad_summaries)
# Output directory for models and summaries
timestamp = str(int(time.time()))
out_dir = os.path.abspath(os.path.join(os.path.curdir, "runs", timestamp))
print("Writing to {}\n".format(out_dir))
# Summaries for loss and accuracy
loss_summary = tf.scalar_summary("loss", cnn.loss)
acc_summary = tf.scalar_summary("accuracy", cnn.accuracy)
f1_summary = tf.scalar_summary("avg_f1", cnn.avg_f1)
# Train Summaries
train_summary_op = tf.merge_summary([loss_summary, acc_summary, f1_summary, grad_summaries_merged])
train_summary_dir = os.path.join(out_dir, "summaries", "train")
train_summary_writer = tf.train.SummaryWriter(train_summary_dir, sess.graph_def)
# Dev summaries
dev_summary_op = tf.merge_summary([loss_summary, acc_summary, f1_summary])
dev_summary_dir = os.path.join(out_dir, "summaries", "dev")
dev_summary_writer = tf.train.SummaryWriter(dev_summary_dir, sess.graph_def)
# Test summaries
test_summary_op = tf.merge_summary([loss_summary, acc_summary, f1_summary])
test_summary_dir = os.path.join(out_dir, "summaries", "test")
test_summary_writer = tf.train.SummaryWriter(test_summary_dir, sess.graph_def)
# Checkpoint directory. Tensorflow assumes this directory already exists so we need to create it
checkpoint_dir = os.path.abspath(os.path.join(out_dir, "checkpoints"))
checkpoint_prefix = os.path.join(checkpoint_dir, "model")
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
saver = tf.train.Saver(tf.all_variables())
# Initialize all variables
sess.run(tf.initialize_all_variables())
def train_step(x_batch, y_batch, x_batch_lex):
"""
A single training step
"""
feed_dict = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
# lexicon
cnn.input_x_lexicon: x_batch_lex,
cnn.dropout_keep_prob: FLAGS.dropout_keep_prob
}
_, step, summaries, loss, accuracy, neg_r, neg_p, f1_neg, f1_pos, avg_f1, hh, hhl = sess.run(
[train_op, global_step, train_summary_op, cnn.loss, cnn.accuracy,
cnn.neg_r, cnn.neg_p, cnn.f1_neg, cnn.f1_pos, cnn.avg_f1, cnn.h_lex_list[0], cnn.h_list[0]],
feed_dict)
time_str = datetime.datetime.now().isoformat()
# print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
print("{}: step {}, loss {:g}, acc {:g}, neg_r {:g} neg_p {:g} f1_neg {:g}, f1_pos {:g}, f1 {:g}".
format(time_str, step, loss, accuracy, neg_r, neg_p, f1_neg, f1_pos, avg_f1))
train_summary_writer.add_summary(summaries, step)
def dev_step(x_batch, y_batch, x_batch_lex, writer=None):
"""
Evaluates model on a dev set
"""
feed_dict = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
# lexicon
cnn.input_x_lexicon: x_batch_lex,
cnn.dropout_keep_prob: 1.0
}
step, summaries, loss, accuracy, neg_r, neg_p, f1_neg, f1_pos, avg_f1 = sess.run(
[global_step, dev_summary_op, cnn.loss, cnn.accuracy,
cnn.neg_r, cnn.neg_p, cnn.f1_neg, cnn.f1_pos, cnn.avg_f1],
feed_dict)
time_str = datetime.datetime.now().isoformat()
print("{}: step {}, loss {:g}, acc {:g}, neg_r {:g} neg_p {:g} f1_neg {:g}, f1_pos {:g}, f1 {:g}".
format(time_str, step, loss, accuracy, neg_r, neg_p, f1_neg, f1_pos, avg_f1))
if writer:
writer.add_summary(summaries, step)
return avg_f1
# if avg_f1>56:
# return True
#
# else:
# return False
def test_step(x_batch, y_batch, x_batch_lex, writer=None):
"""
Evaluates model on a test set
"""
feed_dict = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
# lexicon
cnn.input_x_lexicon: x_batch_lex,
cnn.dropout_keep_prob: 1.0
}
step, summaries, loss, accuracy, neg_r, neg_p, f1_neg, f1_pos, avg_f1 = sess.run(
[global_step, dev_summary_op, cnn.loss, cnn.accuracy,
cnn.neg_r, cnn.neg_p, cnn.f1_neg, cnn.f1_pos, cnn.avg_f1],
feed_dict)
time_str = datetime.datetime.now().isoformat()
print("{}: step {}, loss {:g}, acc {:g}, neg_r {:g} neg_p {:g} f1_neg {:g}, f1_pos {:g}, f1 {:g}".
format(time_str, step, loss, accuracy, neg_r, neg_p, f1_neg, f1_pos, avg_f1))
if writer:
writer.add_summary(summaries, step)
# if avg_f1 > 56:
# return True
#
# else:
# return False
return avg_f1
# Generate batches
batches = cnn_data_helpers.batch_iter(
list(zip(x_train, y_train, x_lex_train)), FLAGS.batch_size, FLAGS.num_epochs)
# Training loop. For each batch...
for batch in batches:
x_batch, y_batch, x_batch_lex = zip(*batch)
train_step(x_batch, y_batch, x_batch_lex)
current_step = tf.train.global_step(sess, global_step)
if current_step % FLAGS.evaluate_every == 0:
print("\nEvaluation:")
curr_af1_dev = dev_step(x_dev, y_dev, x_lex_dev, writer=dev_summary_writer)
# path = saver.save(sess, checkpoint_prefix, global_step=current_step)
# print("Saved model checkpoint to {}\n".format(path))
print("")
print("\nTest:")
curr_af1_tst = test_step(x_test, y_test, x_lex_test, writer=test_summary_writer)
# path = saver.save(sess, checkpoint_prefix, global_step=current_step)
# print("Saved model checkpoint to {}\n".format(path))
print("")
if curr_af1_dev > max_af1_dev:
max_af1_dev = curr_af1_dev
index_at_max_af1_dev = current_step
af1_tst_at_max_af1_dev = curr_af1_tst
print '\nStatus:\n[%d] Max f1 for dev (%f), Max f1 for tst (%f)\n' % (
index_at_max_af1_dev, max_af1_dev, af1_tst_at_max_af1_dev)
sys.stdout.flush()
dictionary["<PAD/>"] = default_vector
# models.append(dictionary)
return norm_model, raw_model
with Timer("lex"):
norm_model, raw_model = load_lexicon_unigram()
with Timer("w2v"):
w2vmodel = load_w2v()
unigram_lexicon_model = raw_model
# unigram_lexicon_model = raw_model
x_train, y_train, x_lex_train = cnn_data_helpers.load_data('trn',w2vmodel, unigram_lexicon_model, max_len)
x_dev, y_dev, x_lex_dev = cnn_data_helpers.load_data('dev', w2vmodel, unigram_lexicon_model, max_len)
x_test, y_test, x_lex_test = cnn_data_helpers.load_data('tst', w2vmodel, unigram_lexicon_model, max_len)
# x_train, y_train = cnn_data_helpers.load_data('trn',w2vmodel , max_len)
# x_dev, y_dev = cnn_data_helpers.load_data('dev', w2vmodel, max_len)
# x_test, y_test = cnn_data_helpers.load_data('tst', w2vmodel, max_len)
del(w2vmodel)
del(norm_model)
del(raw_model)
gc.collect()
print("Train/Dev split: {:d}/{:d}".format(len(y_train), len(y_dev)))
def run_train(w2vdim, w2vnumfilters, lexdim, lexnumfilters, randomseed,
withlexicon):
max_len = 60
with Timer("lex"):
norm_model, raw_model = load_lexicon_unigram(lexdim)
with Timer("w2v"):
w2vmodel = load_w2v(w2vdim)
unigram_lexicon_model = norm_model
# unigram_lexicon_model = raw_model
x_train, y_train, x_lex_train = cnn_data_helpers.load_data(
'trn', w2vmodel, unigram_lexicon_model, max_len)
x_dev, y_dev, x_lex_dev = cnn_data_helpers.load_data(
'dev', w2vmodel, unigram_lexicon_model, max_len)
x_test, y_test, x_lex_test = cnn_data_helpers.load_data(
'tst', w2vmodel, unigram_lexicon_model, max_len)
del (w2vmodel)
del (norm_model)
del (raw_model)
gc.collect()
print("Train/Dev split: {:d}/{:d}".format(len(y_train), len(y_dev)))
# Training
# ==================================================
if randomseed > 0:
tf.set_random_seed(randomseed)
with tf.Graph().as_default():
max_af1_dev = 0
index_at_max_af1_dev = 0
af1_tst_at_max_af1_dev = 0
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default():
if randomseed > 0:
tf.set_random_seed(randomseed)
cnn = TextCNN(sequence_length=x_train.shape[1],
num_classes=3,
embedding_size=w2vdim,
embedding_size_lex=lexdim,
lex_filter_size=lexnumfilters,
filter_sizes=list(
map(int, FLAGS.filter_sizes.split(","))),
num_filters=w2vnumfilters,
with_lexicon=withlexicon,
l2_reg_lambda=FLAGS.l2_reg_lambda)
# Define Training procedure
global_step = tf.Variable(0, name="global_step", trainable=False)
optimizer = tf.train.AdamOptimizer(1e-3)
grads_and_vars = optimizer.compute_gradients(cnn.loss)
train_op = optimizer.apply_gradients(grads_and_vars,
global_step=global_step)
# Keep track of gradient values and sparsity (optional)
grad_summaries = []
for g, v in grads_and_vars:
if g is not None:
grad_hist_summary = tf.histogram_summary(
"{}/grad/hist".format(v.name), g)
sparsity_summary = tf.scalar_summary(
"{}/grad/sparsity".format(v.name),
tf.nn.zero_fraction(g))
grad_summaries.append(grad_hist_summary)
grad_summaries.append(sparsity_summary)
grad_summaries_merged = tf.merge_summary(grad_summaries)
# Output directory for models and summaries
timestamp = str(int(time.time()))
out_dir = os.path.abspath(
os.path.join(os.path.curdir, "runs", timestamp))
print("Writing to {}\n".format(out_dir))
# Summaries for loss and accuracy
loss_summary = tf.scalar_summary("loss", cnn.loss)
acc_summary = tf.scalar_summary("accuracy", cnn.accuracy)
f1_summary = tf.scalar_summary("avg_f1", cnn.avg_f1)
# Train Summaries
train_summary_op = tf.merge_summary(
[loss_summary, acc_summary, f1_summary, grad_summaries_merged])
train_summary_dir = os.path.join(out_dir, "summaries", "train")
train_summary_writer = tf.train.SummaryWriter(
train_summary_dir, sess.graph_def)
# Dev summaries
dev_summary_op = tf.merge_summary(
[loss_summary, acc_summary, f1_summary])
dev_summary_dir = os.path.join(out_dir, "summaries", "dev")
dev_summary_writer = tf.train.SummaryWriter(
dev_summary_dir, sess.graph_def)
# Test summaries
test_summary_op = tf.merge_summary(
[loss_summary, acc_summary, f1_summary])
test_summary_dir = os.path.join(out_dir, "summaries", "test")
test_summary_writer = tf.train.SummaryWriter(
test_summary_dir, sess.graph_def)
# Checkpoint directory. Tensorflow assumes this directory already exists so we need to create it
checkpoint_dir = os.path.abspath(
os.path.join(out_dir, "checkpoints"))
checkpoint_prefix = os.path.join(checkpoint_dir, "model")
if not os.path.exists(checkpoint_dir):
os.makedirs(checkpoint_dir)
saver = tf.train.Saver(tf.all_variables())
# Initialize all variables
sess.run(tf.initialize_all_variables())
def train_step(x_batch, y_batch, x_batch_lex):
"""
A single training step
"""
feed_dict = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
# lexicon
cnn.input_x_lexicon: x_batch_lex,
cnn.dropout_keep_prob: FLAGS.dropout_keep_prob
}
_, step, summaries, loss, accuracy, neg_r, neg_p, f1_neg, f1_pos, avg_f1 = sess.run(
[
train_op, global_step, train_summary_op, cnn.loss,
cnn.accuracy, cnn.neg_r, cnn.neg_p, cnn.f1_neg,
cnn.f1_pos, cnn.avg_f1
], feed_dict)
time_str = datetime.datetime.now().isoformat()
# print("{}: step {}, loss {:g}, acc {:g}".format(time_str, step, loss, accuracy))
#print("{}: step {}, loss {:g}, acc {:g}, neg_r {:g} neg_p {:g} f1_neg {:g}, f1_pos {:g}, f1 {:g}".
# format(time_str, step, loss, accuracy, neg_r, neg_p, f1_neg, f1_pos, avg_f1))
train_summary_writer.add_summary(summaries, step)
def dev_step(x_batch, y_batch, x_batch_lex, writer=None):
"""
Evaluates model on a dev set
"""
feed_dict = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
# lexicon
cnn.input_x_lexicon: x_batch_lex,
cnn.dropout_keep_prob: 1.0
}
step, summaries, loss, accuracy, neg_r, neg_p, f1_neg, f1_pos, avg_f1 = sess.run(
[
global_step, dev_summary_op, cnn.loss, cnn.accuracy,
cnn.neg_r, cnn.neg_p, cnn.f1_neg, cnn.f1_pos,
cnn.avg_f1
], feed_dict)
time_str = datetime.datetime.now().isoformat()
print(
"{} : {} step {}, loss {:g}, acc {:g}, neg_r {:g} neg_p {:g} f1_neg {:g}, f1_pos {:g}, f1 {:g}"
.format("DEV", time_str, step, loss, accuracy, neg_r,
neg_p, f1_neg, f1_pos, avg_f1))
if writer:
writer.add_summary(summaries, step)
return avg_f1
def test_step(x_batch, y_batch, x_batch_lex, writer=None):
"""
Evaluates model on a test set
"""
feed_dict = {
cnn.input_x: x_batch,
cnn.input_y: y_batch,
# lexicon
cnn.input_x_lexicon: x_batch_lex,
cnn.dropout_keep_prob: 1.0
}
step, summaries, loss, accuracy, neg_r, neg_p, f1_neg, f1_pos, avg_f1 = sess.run(
[
global_step, dev_summary_op, cnn.loss, cnn.accuracy,
cnn.neg_r, cnn.neg_p, cnn.f1_neg, cnn.f1_pos,
cnn.avg_f1
], feed_dict)
time_str = datetime.datetime.now().isoformat()
print(
"{} : {} step {}, loss {:g}, acc {:g}, neg_r {:g} neg_p {:g} f1_neg {:g}, f1_pos {:g}, f1 {:g}"
.format("TEST", time_str, step, loss, accuracy, neg_r,
neg_p, f1_neg, f1_pos, avg_f1))
if writer:
writer.add_summary(summaries, step)
return avg_f1
# Generate batches
batches = cnn_data_helpers.batch_iter(
list(zip(x_train, y_train, x_lex_train)), FLAGS.batch_size,
FLAGS.num_epochs)
# Training loop. For each batch...
for batch in batches:
x_batch, y_batch, x_batch_lex = zip(*batch)
train_step(x_batch, y_batch, x_batch_lex)
current_step = tf.train.global_step(sess, global_step)
if current_step % FLAGS.evaluate_every == 0:
print("Evaluation:")
curr_af1_dev = dev_step(x_dev,
y_dev,
x_lex_dev,
writer=dev_summary_writer)
# path = saver.save(sess, checkpoint_prefix, global_step=current_step)
# print("Saved model checkpoint to {}\n".format(path))
curr_af1_tst = test_step(x_test,
y_test,
x_lex_test,
writer=test_summary_writer)
# path = saver.save(sess, checkpoint_prefix, global_step=current_step)
# print("Saved model checkpoint to {}\n".format(path))
if curr_af1_dev > max_af1_dev:
max_af1_dev = curr_af1_dev
index_at_max_af1_dev = current_step
af1_tst_at_max_af1_dev = curr_af1_tst
print 'Status: [%d] Max f1 for dev (%f), Max f1 for tst (%f)\n' % (
index_at_max_af1_dev, max_af1_dev,
af1_tst_at_max_af1_dev)
sys.stdout.flush()
if self.name:
print '[%s]' % self.name,
print 'Elapsed: %s' % (time.time() - self.tstart)
def load_w2v():
model_path = '/Users/bong/works/data/word2vec_twitter_model/word2vec_twitter_model.bin'
with Timer("load w2v"):
model = Word2Vec.load_word2vec_format(model_path, binary=True)
print("The vocabulary size is: " + str(len(model.vocab)))
return model
w2vmodel = load_w2v()
x_train, y_train = cnn_data_helpers.load_data('trn', w2vmodel, max_len)
x_dev, y_dev = cnn_data_helpers.load_data('dev', w2vmodel, max_len)
# x_test, y_test = cnn_data_helpers.load_data('tst', w2vmodel, max_len)
del (w2vmodel)
gc.collect()
print("Train/Dev split: {:d}/{:d}".format(len(y_train), len(y_dev)))
# Training
# ==================================================
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
keys = raw_model[index].keys()
dictionary = dict(zip(keys, new_val))
norm_model[index] = dictionary
return norm_model, raw_model
def load_w2v():
return {}
norm_model, raw_model = load_lexicon_unigram(14)
w2vmodel = load_w2v()
with Timer("load dataset"):
x_test, y_test, x_lex_test = cnn_data_helpers.load_data('dev', w2vmodel, norm_model, 60)
with tf.Graph().as_default():
sess = tf.Session()
with sess.as_default():
cnn = TextCNN(
sequence_length=60,
num_classes=3,
embedding_size=400,
embedding_size_lex=14,
lex_filter_size=9,
filter_sizes=list(map(int, ['2','3','4','5'])),
num_filters=256,
print 'Elapsed: %s' % (time.time() - self.tstart)
def load_w2v():
model_path = '/Users/bong/works/data/word2vec_twitter_model/word2vec_twitter_model.bin'
with Timer("load w2v"):
model = Word2Vec.load_word2vec_format(model_path, binary=True)
print("The vocabulary size is: " + str(len(model.vocab)))
return model
w2vmodel = load_w2v()
# x_train, y_train = cnn_data_helpers.load_data('trn',w2vmodel , max_len)
# x_dev, y_dev = cnn_data_helpers.load_data('dev', w2vmodel, max_len)
x_test, y_test = cnn_data_helpers.load_data('tst', w2vmodel, max_len)
del (w2vmodel)
gc.collect()
# savepath = '/Users/bong/works/tfen/w2v_cnn/runs/backup/checkpoints/model-2900'
savepath = '/Users/bong/works/tfen/w2v_cnn/runs/1464326614/checkpoints/model-6700'
with tf.Graph().as_default():
session_conf = tf.ConfigProto(
allow_soft_placement=FLAGS.allow_soft_placement,
log_device_placement=FLAGS.log_device_placement)
sess = tf.Session(config=session_conf)
with sess.as_default():
cnn = TextCNN(sequence_length=x_test.shape[1],
num_classes=3,